Human Brain Still Beats Computers At Finding Messages and Meaning Within Noise (hackaday.com)

szczys writes: One thing the human brain still does a lot better than computers is to recognize patterns within noise. That's why CAPTCHA uses distorted images to prove you're human, and random number generators are often inspected by visual representation. There is a technology that leverages this human knack for signal processing to make us part of the machine. The Hellschreiber is a communications device which has no idea whatsoever what the message actually is. It transfers a signal from one unit to the next, before being assembled into an image. A human looking at the image will see words, much like CAPTCHA. But even if the signal isn't perfect, our brains can often pick out the order within the madness, much like inspecting a PRNG for uniform distribution.

Comments

Are The Computers Better

[Greyfox]

How about at finding messages that aren't there in the noise? The human brain is excellent at doing that, so in the end the computer might win on error rate.

Re:Are The Computers Better

[lhowaf]

Paul is dead.

Re:Are The Computers Better

[Spy Handler]

Wrong. I saw lizards in photographs that Mars Rover totally missed.

Re:Are The Computers Better

[zAPPzAPP]

That would be because the information is hidden on a level, frequency, or timescale where our brains can not detect it.
Not a fair comparison.

Re:Are The Computers Better

[thegarbz]

I would bet against you with a large sum. The only way a computer would know this is if it could identify what part of the pattern constitutes the encoded data and what part constitutes the background "noise". For that it would need to know what noise looks like, and what the pattern looks like.

To extend this with an example. If you rename a JPEG without an extension you would need to find some evidence of a file being a JPEG in the first place. It won't get this from the raw data stream unless it is looking and comparing for an exact signature of a JPEG file and even them I'm sceptical if it could identify the encoding without any metadata as the computer would need to know what a picture of a kitten looks like, compared to say random noise it would get on a failed attempt to decode. By extension to the other side of the problem, even if I AM A JPEG was expressly written in the metadata that could be easily hidden within the audio file and a computer would never identify it because it would need to know what is audio and what is not.

Computers can only look for structured patterns these patterns need to be defined by a person up front. This is how computers can look for words and get creepy meaning from the bible proving that the world will end on a day of the operator's choosing. But you can't feed it a song and have it spit back at you: There's a JPEG in here, without feeding it a known clean song to compare.

Re:Are The Computers Better

[The-Ixian]

I AM A JPEG

Best. Costume. Idea. Ever.

pareidolia

[aepervius]

We human excel at finding specific shapes, like a human shape or face. Change the shape , like to number and letter , and we are far less good as show the problem people also have with captcha. Get a random shape, and computer will *excel* way over human to decide where it should go. So I would argue that our "finding shape out of random noise" is actually much more specialized than surmised.

Hey /. stop logging me out :(

[Chuffpole]

It's no good giving me mod points every couple of days hoping I'll moderate, if you keep logging me out all the time!

I'm logged in on the home page, but when I open stories in separate tabs most of them are logged out. Developers.Slashdot is one of the few that still seems OK.

When you're done messing about with breaking things on older Firefoxes (yes I'm on v3 on Mint 7 but no reason to upgrade when it mostly still works), please fix this. Thanks. Happy 2016.